-
-
Notifications
You must be signed in to change notification settings - Fork 906
OpenAug 17, 2026
No due date
•Last updated 0% complete
List view
0 of 2 selected 0 issues of 2 selected
Adam/AdamW/LAMB/AdEMAMix: weight decay applied in wrong order in default and Triton backends, diverging from CUDA kernel
CUDAIssues and PRs related to the CUDA backend, excluding installation/support help.Issues and PRs related to the CUDA backend, excluding installation/support help.OptimizersIssues or feature requests relating to optimizersIssues or feature requests relating to optimizersStatus: Open.Fix decoupled weight decay ordering in the CUDA Adam/AdEMAMix kernels
CUDAIssues and PRs related to the CUDA backend, excluding installation/support help.Issues and PRs related to the CUDA backend, excluding installation/support help.OptimizersIssues or feature requests relating to optimizersIssues or feature requests relating to optimizersStatus: Open (in progress).bitsandbytes-foundation/bitsandbytesnumber 2040#2040 In bitsandbytes-foundation/bitsandbytes;